reservoirpy.ops.link_feedback#

class reservoirpy.ops.link_feedback(node, feedback, inplace=False, name=None)[source]#

Create a feedback connection between the feedback node and node. Feedbacks nodes will be called at runtime using data from the previous call.

This is not an in-place operation by default. This function will copy node and then sets the copy _feedback attribute as a reference to feedback node. If inplace is set to True, then node is not copied and the feedback is directly connected to node. If feedback is a list of nodes or models, then all nodes in the list are first connected to a Concat node to create a model gathering all data from all nodes in a single feedback vector.

You can also perform this operation using the << operator:

node1 = node1 << node2
# with feedback from a Model
node1 = node1 << (fbnode1 >> fbnode2)
# with feedback from a list of nodes or models
node1 = node1 << [fbnode1, fbnode2, ...]

Which means that a feedback connection is now created between node1 and node2. In other words, the forward function of node1 depends on the previous output of node2:

\[\mathrm{node1}(x_t) = \mathrm{node1}(x_t, \mathrm{node2}(x_{t - 1}))\]

You can also use this function to define feedback:

node1 = link_feedback(node1, node2)
# without copy (node1 is the same object throughout)
node1 = link_feedback(node1, node2, inplace=True, name="n1_copy")

Parameters:

node (Node) – Node receiving feedback.
feedback (_Node) – Node or Model sending feedback
inplace (bool, defaults to False) – If True, then the function returns a copy of node.
name (str, optional) – Name of the copy of node if inplace is True.

Returns:

A node instance taking feedback from feedback.

Return type:

Node

Raises:

TypeError –

If node is a Model. Models can not receive feedback. - If any of the feedback nodes are not _Node instances.